MAPPS kicked off the year in January by hosting their annual winter meetings, this year in beautiful and warm Hawaii. During the meetings the winners of the Geospatial Excellence Awards were announced. Aerial Services, Inc. won the MAPPS Geospatial Excellence Award for the fourth time. The submission, “Bending Space and Time”, won the excellence award for the GIS/IT/Remote Sensing category. There were two parts to the project entry, the write-up and the creative. Take a closer look and let us know what you think about our project!
The Creative: The project from a 30,000′ view is about bringing historical scanning to life so the poster itself is a film scan with true fiducial and other identifier markings. The poster also creates a unique visual which demonstrates the process of taking a flat 60+ year old film scan and draping it over current DEM maps and shows the vast territory in which ASI was able to produce data for.
CHECK OUT THE POSTER BELOW
The write-up review: ASI was tasked with georeferencing historical imagery for 59 counties in South Dakota with the goal to produce 59 countywide mosaics. During the process ASI discovered several issues that would lead them to pioneer a new process which enabled cost-effective georeferencing of historical imagery and exceeded accuracy requirements by 2x. With this new process we can gain a better understanding of our land and environmental and economic impacts for a diverse range of applications.
READ FULL WRITE-UP BELOW
Project and client background
The United States Department of Agriculture (USDA) Aerial Photography Field Office (APFO) is a division of the Farm Service Agency (FSA). Among other things, it provides georeferenced scans of historic aerial photography. Georeferenced scans provide a variety of geospatial services for the public which can be used for a wide variety of purposes.
The APFO has directed the acquisition, use and distribution of aerial photography for more than 65 years. Several years ago, the APFO began scanning the aerial photography to preserve and more efficiently store this source of immense geospatial information. However, they had the foresight to understand that a vast collection of scanned imagery is far more useful to their customers when its geospatial relevance is intact.
Project Objectives: In the beginning of 2018, Aerial Services, Inc. (ASI) was contracted to assist the AFPO in achieving two main goals:
- Spatially enable historical imagery datasets, and
- Create modern day access to historical agriculture imagery.
ASI was awarded a task order to accurately georeference and create mosaics of approximately 51,400 scans of aerial photography originally acquired in 1967, 1969, 1971, 1974, and 1976. All photos were acquired in South Dakota. Requirements stated that these scans must be processed as digital orthophoto quarter quads (DOQQs) and assembled to make 59 countywide mosaics.
Intended Deliverable Use(s): The intended use of the DOQQs and county-wide mosaics was to provide the NRCS rich geospatial information for planning and compliance activities, which support provisions of the Food Security Act. This information would be made available to the public and enable many other diverse uses.
What was supplied: The original aerial photography was acquired in remote sensing projects conducted between 1967 and1976. The film scans, provided by APFO, were uncompressed grayscale TIFF images (9” x 9” aerial film) scanned at 12.5 microns (330 megapixels for each scan). Shapefiles were supplied by the APFO, representing the photo-index of the approximate center point of each aerial photograph. Also supplied to ASI, were compressed county mosaics (CCMs) created from modern-day National Aerial Imagery Program (NAIP) photography. The CCMs were used as the basemap to which the scans had to be accurately georeferenced.
The process: ASI accurately georeferenced the photo exposure stations in order to orthorectify and assemble the frames into mosaics. This process was broken into two main actions.
Step 1. The first step was to determine the interior and exterior parameters of the camera and exposure stations for each aerial frame scan through aerotriangulation. This iterative process involved determining the relative orientations of bundles of frames and selecting ground control points (GCP’s) using a variety of historic features still visible in more current CCM’s. After a sufficient number of GCPs were measured and visible in both the historic scans and the CCMs, an elevation coordinate was assigned to each GCP by interpolating the elevation from a modern DEM. Historical frames that could not achieve an accurate relative orientation or did not align with ground control to acceptable accuracy requirements had to be manually adjusted to ensure high accuracy of the final product.
Step 2. After camera/lens parameters and external orientations were calculated, the frames were orthorectified to the modern DEM and then seamed together. Seamline mismatches and anomalies to the CCM’s were noted and corrections performed where necessary. After the final mosaic was assembled, technicians evaluated “fit” of all areas of the mosaic to the reference CCM. Positional accuracy of the mosaic was tested using National Standard for Spatial Data Accuracy (NSSDA) reporting methods. The final mosaic was then cut into DOQQ tiles each covering approximately 15 square miles.
Expected Deliverables: Delivery of the first county project was due within 30 days of task order award. ASI successfully met all schedules specified by project requirements. DOQQs were delivered on external hard drives to the USDA APFO. Also included were shapefiles showing the multiple seamlines between all scanned aerial frames within each DOQQ, FGDC-compliant metadata files, and a NSSDA accuracy report.
Major Accomplishments: USDA APFO teaming with ASI were able to meet every project deadline and exceed project accuracy standards to achieve the main project objectives.
- Enabled the cost-effective georeferencing of 60-year-old flat imagery scans.
- Produced a rich source of 50+ year old spatial information, covering 59 US counties spanning several years, resulting in the data being far more accessible to modern GIS technology.
- ASI well-exceeded the high positional accuracy standards required by the client.
There were a number of difficulties aligning 50 to 60-year-old aerial photography with current aerial imagery and elevation models. The age of the film, inaccuracy of older camera systems, and changes to natural and man-made landscapes all contribute to make accurately georeferencing this historical imagery problematic. ASI designed a workflow to manage these complexities and ensure the delivery of a product that exceeded the positional accuracy requirements within a fixed budget. This workflow development was a challenge that required considerable ingenuity.
Problem 1: ASI’s traditional georeferencing software was unable to properly aero-triangulate a sufficient number of scans where the image content was too uniform in appearance, such as heavily forested and certain agricultural areas. In some counties, a large number of scans may have this type of image content. These areas required an efficient method of mathematically aligning the adjacent frames.
Problem 2: Many frames and the frame-center coordinates provided by the client were erroneous. These frames could not be georeferenced to the correct location within a county without first identifying that they had bad coordinates. It was necessary to correct the coordinates so the frames could align with others in their vicinity.
Innovative Solution: Our experienced team of geospatial professionals engineered a solution that addressed both of these issues by utilizing new COTS Structure From Motion (SFM) software that had not previously been used for this purpose. (This will be explained in greater length in the Innovative Application Development section.)
The final challenge our team faced was the positional accuracy required for final deliverables from this 60-year-old historical imagery. Although the innovative solution developed for this project was selected because it was a solution to the first two challenges presented, it also produced better accuracy than initially anticipated. It was more than twice as accurate as required. That is an extremely beneficial outcome of our innovative methods considering the USDA APFO’s clients use case for this geospatial information. Environmental assessments, change detection, past land use analysis, and property boundary disputes depend upon solid positional accuracy.
The initial accuracy that was required to exceed 10 meters. ASI’s deliverables established a final accuracy of 3.6 meters using NSSDA reporting.
Original or Innovative Application Development
ASI’s initial plan presented problems (discussed above) and it became immediately clear a new plan was needed. ASI’s new innovative workflow was able to overcome two main hurdles presented by the project; our original software’s inability to properly aero-triangulate the scans accurately and its inability to position photos with missing, incorrect, or inaccurate centerpoints. Both of these issues were a major hindrance for our team and for the USDA APFO.
ASI’s solution was to utilize structure from motion (SFM) software. This technology was developed and popularized for aerial drone remote sensing and surveying and provided several technological advantages.
First, the SFM algorithms can find, with very few known parameters, many significant pass and tie points within images. The software can accurately align blocks of hundreds or thousands of photos using tens of thousands of pass and tie points per image pair. These pass and tie points were even found in areas of uniform content where it was very difficult or impossible for other software to find. It was also significantly faster than a human operator to manually locate and digitize points. SFM software is also able to perform the aerotriangulation processes in minutes, matching or beating the performance of other software.
The strength of SFM software is also clear in forested or vegetation-barren areas where control points would be nearly impossible to relate to imagery with 50+ years difference. Control points near but outside of these areas allow for accurate triangulation even when matches are not found in current and historic imagery. A typical county of 500 frames would need only approximately 20 control points to achieve better than 5m accuracy.
Second, while traditional aerotriangulation software relies on relatively accurate positions and orientations of photos, SFM algorithms work well with data where little to nothing is known about the origin of the photo. The SFM software allowed the placement and aerotriangulation of photos with poor or no geolocation at all. With ground-based frame center estimates based on previous matching of single 2D features from the frame to the ground, any roll or pitch in the photo would cause an error in the airborne frame station position estimate. In addition, many of the frame centers were interpolated, causing errors from non-linear flight paths and inconsistent camera triggering. Elevations were difficult to estimate due to the collection altitude only being estimated by the film scale. It is common for single frame center estimates prior to aerotriangulation to have errors of several hundred meters in X, Y, and Z. While other software requires a somewhat accurate initial position estimate, drone mapping SFM software excels at finding image stations based on image similarity. In addition, SFM software can relocate an incorrectly estimated photo center across a project. The SFM software is able to quickly and accurately generate relative internal and external orientations of photos without any control or previous geolocation in a relative alignment. Prior to adding GCP’s, SFM software is able to align these frames and increase accuracy to tens of meters.
Third, SFM software could quickly and accurately estimate internal camera and lens parameters via in-situ calibration for every group of frames. Often the majority of the frames did not contain calibration data including a calibrated focal length. An external process of calibrating each camera would take significant time and resources, however the SFM software conducted the estimation using data from thousands of frames.
Fourth, an all-in-one drone SFM program has the ability to autocorrelate its own Digital Elevation Model (DEM). In some cases, inaccuracy is caused by the orthorectification process when the imagery is orthorectified or “warped” to a DEM. 60-year-old terrain does not always match the shape of today’s DEMs in the same areas. When significant erosion or earth movements occur in these areas, or when frames do not meet global vertical accuracy, a new surface must be generated to maintain 2D accuracy when orthorectifying the historical images. Drone SFM software can generate an autocorrelated surface by matching pixels. These point clouds and resultant surfaces are often quite noisy are require further processing. Modern LiDAR algorithms can clean the data and generate a ground surface useful for accurate orthorectification without smears or displacement.
Future Value to Geospatial Profession and the Public
The USDA’s historical aerial photography plays a vital role in environmental assessments, change detection, past land use analysis, and property boundary disputes. Their customers include attorneys, universities, local governments, Federal Agencies, and individuals.
The development of this innovative and highly efficient process of creating elevation models and orthomosaics from existing historical images has opened the door to a great deal of data which was previously unknown or impracticable to process with other methods.
Soil erosion and flooding have major impacts on agriculture, coastal communities, and wildlife ecosystems across the country. In the three previous years, the average insurance claims for flooding in the United States is $1.9 billion per year. By using this new process pioneered by ASI, we are now able to more efficiently and cost-effectively produce imagery with substantially better accuracy and enable this historical information to be easily accessible using any modern GIS system. With access to this rich, new geospatial data, we will gain a better understanding of our land and have large environmental and economic impacts for a diverse range of projects and groups.
Client / Owner Satisfaction
ASI’s has maintained a long-standing partnership with the USDA to complete the yearly historical scanning project. Due to ASI’s ability to meet deadlines and provide unmatched accuracy with the state of South Dakota, ASI was awarded the entire state of Alabama’s task order.
With an increasing number of applications and customers for historical imagery, the USDA has increasingly depended on ASI’s team to georectify and modernize their data library. As the demand for georectified historical data increases, the value offered by services like ASI’s with continue to increase. This is the biggest testament to the USDA’s satisfaction with ASI’s contribution to this project. During this project, all project schedules were met and all accuracy standards were exceeded.
In the end, the USDA was seeking ways to spatially enable their datasets, create modern day access to historical imagery, and leverage new technologies resulting in easier viewing, analysis, and delivery methods for their customers. ASI listened, analyzed, developed and in the end, overcame multiple obstacles. ASI met delivery times and exceeded accuracy standards, leaving the client extremely satisfied.